N E , T H E R L A N D S G E , O D E , T I CC O M M I S S I O N

P U B L I C A T I O N SO N G E O D E S Y
VOLUME 5

NEW SERIES
NUMBER I

S-TRANSF'ORMATIONS
AND CRITERION MATRICES
by
W. BAARDA
Computing Centre of the Delft Geodetic Institute
Delft Universityof Technology

S E C O N DR E V I S E DE D I T I O N I 9 8 I

, H I J S S E W E G1 I , D E L F T . N E T H E R L A N D S
R I J K S C O M M I S S I EV O O R G E O D E S I E T

S.TRANSFORMATIONS AND CRITERION MATRICES

l1

well as possible {in the latter case it will usually be termed "substitute matrix"}. Testing
or description turns out to be possible only within one S-system; this implies that the
theory of S-transformations is indispensable.In section l7 it is demonstrated, by means
of a practical method of network computation, why these transformations could escape
notice, with the consequencesfor the interpretation of the results obtained.
Having in mind the example of levelling networks in section ll, it is tried in the second
part to explain the assumption, in section 72, of the optional function d!,. ln the construction of the criterion matrix, it is introduced as a positive "polynomial" {better: posynomial}
of the distance /,, between the points P, and P; of the network. It is sufficient to compute
/,, from approximate coordinates in a limited number of significant digits. In section 13,
the variance Adi is introduced, describing the uncertainty of definition of a terrain point
P, as a mathematical point. In the note to section 13, the requirements to be fulfilled by
the optional function d,2,are investigated.
The third part presents the generalization to two-dimensional coordinate variates; like
in the secondpart, the construction of criterion matricesis grafted upon S-transformations.
Becauseone is dealing with a genuine complex number theory, some small parts of the
calculus of observations had to be treated by complex numbers in section 14. Among
them are treated the consequencesof the requirement of circular point- and relative standard ellipsesin the construction of the criterion matrix. The construction itself is developed
in section 15, supplementedby a different approach to S-transformations in section 16,
and a later development in section l5a. After ample consideration,both coursesof development have been maintained and presentedtogether, in order to show the reader different
aspectsof the theory. The course followed in section l5 dates from1962-1963; eventually
it led via the G-functions to an optional function d,2,of the sametype as the one mentioned
above. Formulas for the computation of the criterion matrix proved, however, to be
subjectto some inconvenient restrictions. In the summer of 1970,the theory of section l5.a
was therefore developed,which was more directly aimed at the computations,and included
the above mentioned variances ldl . Here, the significance of the parameters of the criterion
matrix is indicated, and a way is sketchedtowards the determination of numerical values
to be assignedto these parameters.Therefore, section l5.a can be consideredas the nucleus
of this paper as far as the construction of the criterion matrix is concerned,the more so
becausethis section - contrary to sections 15 and 16 - is practically unconnected to the
only partially published "Polygon theory in the complex plane" [2]. Finally, in section 17
an adjustment model is developed,as an application of S-transformationsto the connection
ofseveral coordinate systemsbelongingto {stochastically}independentgroups ofcoordinate
variates in different networks. Another aspect of section l7 was already mentioned in the
surnmary of the first part of this paper.
A generalization of the theory to networks of three-dimensional coordinate variates
proves to be possible,but it has not yet been elaborated sufficiently.
The question remains if the choice of the optional function dlrhas been made sufficiently
general, although it is very satisfactorythat for one-, two- and three-dimensionalsituations
the same optional function can be used. This question is open to further research.

S-TRANSFORMATIONS AND CRITERION MATRICES

matrix) must therefore lead to conclusions that are independent of the choice of the
S-system.
Note to section 3. An alternative form of the non-singular S-transformation treated
Take from (3.6) the non-singular transformation:

f,"Y"

-v -v

[tl"*t

-wt

,o

-Dt

,o

'wu

-ow

- v, o - v, o

[t\"-t

lws

-Ds

'\9v

-ow

l*t'"'
1r,",

(3.e)

Lt*"'

and substitutethis into the right-handmemberof (3.7):

(nv")
l

(n:,\
t

riY'l
l n y ,l : r s [ ; l ] , r . 1
1f"9,)

....(3.10)

lf,1*')

Executingthe matrix multiplication concernedthen resultsin:

Eil[ ??)l;'i)

(3.11)

The inversionof (3.11) is then:

l*r"\fu,-+
-,*-+i["'^'l
*:,,1
l
l rr,l:l' o ''"
l
,1ln*')
lnr,) t

12)
.(3

which formula is the equivalentof (2.8), only with the restriction that the transformation
(3.12)givesthe transition of quantitiesin the S-systemto quantitiesin the S-system,
therebyincluding:
l2f;"t

and

/_:\:,u

as "free" variates*)

. (3.13)

*) "Free" observation variates do not occur in condition equations of an adjustment problem, but they
are correlated with {..tied"} variates occurring in theseequations; see [l], section 12.

49

S-TRANSFORMATIONSAND CRITERION MATRTcES

For simplicity, we deviate slightly from the notation in section 13 of [],

the Choleskeydecompositionof H('"):

and write for

(8.7)

1 q , ( r s ) -f - r . f - t

Standardization of rhe /z'o"\-sample spaceis obtained by the transformations:

(/z'u) : f .(/zv'))
(8.8)

- f .(/z'os))
(Az'<'"t1
(8.6) then becomes,with the unit matrix 1:

A . r - t . ( f . G ( ' " ) ' l - *) -, ' 1 1 ' 7 - -' ,r1 -< 0

(s.9)

Now apply an orthogonal transformation, such that the matrix (f 'G('s)'f-) is reduced to
the diagonal matrix:

(8.10)

dgl)"t,"',)",1
henceby the orthogonal transformations:

c.(/:r")

with C-L: C' '

C.(/z,t^t'r,
I
I

' (8'11)

Then (8.9) becomes:

A.f

-tC-r(C.f .6(rs).;*.
- 0
C - _ A .I ) . C * - t . f " * - t . . , { <

or, with (8.10)

0 .
(t.r-t.c-t).ogLLr-1,...,A,-Lf.(A.r-t.c-t)'<

.(8.12',)

dgl)"t- )",...,1,- A]

.(8.I2")

or:
negative semi-definite .

(8.12)is fulfilledby:*)
(8.13)

+) See,e.g., L. Mrnsrv - An Introduction to Linear Algebra - Oxford, 1955, {in particular chapter 13}.

65

S-TRANSFORMATIONS AND CRITERION MATRICES

(12.7) gives a "family" of criterion matrices, every member of this family is characterized
by the choice of the coefficientscr. Not every member is acceptable,a necessarycondition is:

Ux[\,(Axy\

positivedefinite

. (r2.rr)

This question is further investigatedin the note to section 13.

On examining(12.2)and (12.8)it appearsto be possibleto constructa kindof "absolute"
criterion matrix, whose significance is, however, only abstract-theoretical,as shown in
section10. From (12.7) and (12.8)one obtains:
Xi;Xi:

XrrX, -d?i:62

-d?i

and hence:

/_!,

n, I a'
tx,

/_!,

/_t,

@'z-d?j)(d'-d?,)

| @' d3) d2

(r2.12)

@2- d2j,)

n , l @2 -a ? ) @'-d ? ) d 2
Applicationof (12.1)to (12.12)producesagain(12.10).{Seealso(13.22)-(13.25)}'
(12.12)is the type of criterion matrix - with only c, * 0 - as was usedby Beenoe for
eachof the coordinatesx and y separatelyin the treatmentof traversenetworks and area
calculationsin the HTW-1956,and also by Alneno.Lin his investigationin 1963.The
and the author considers
interpretationof d2 in the HTW-1956was not very satisfactory,
its introductionin the presenttheoryas an eliminable"nuisanceparameter"as an elegant
solution for the difficultiesencountered.
and may only be usedto derive(12.10),
Since(12.12)only has a theoreticalsignificance
practical
(l2.ll)
significance;
the discussionabout
positive
has
a
definiteness
and only the
practice.
problems
of
givingnumericalvaluesto d2 is irrelevantfor

S-TRANSFORMATIONS AND CRITERION MATRICES

67

In order to develop the formulae as carefully as possible,/df is assumedto have a different

value for each point P;. In practice one will take an averagevalue for groups of points.
Well-marked points of high order will have a small or negligible value for /d2, this is important for the choice of the base point P, or for the choice of the given points to which
the network is connected in a second step. Therefore the value of /d! will be small as
compared with a value of ldf whenP, is an arbitrary non-marked or poorly marked terrain
point.
The assumption of (13.1) is such, that all relations of section ll are also valid for the
y-variates.
( 1 2 . 1 )b e c o m e s :

/y[n : /\ + /_!i- ax,- /x: : /_i" * /*i - /+|)

/)Y': / *t +/-!3- 4,- /*i : /x'' +/ x:- ^:fl

/y['): /x,+axi-/x,- /xi : g
I

(13.2)

Or, with (13.1):

frrfr : xfnfi+zd?+ad?
+ /d?

vfw:o
y!'),y, : x!'),x1+ad!
(13.3)

Y['),Yi : xfi,xi

nf, y, : tl', t,
:xi

li

,!i

li

,!i

xi ,xi

li

,!,

: xi ;xr

,xi

- /d?
+ /d?

Next,considerthe situationin(12.7),for whichwe obtain,with (13.1)and (13.3):

Y,,Y,-Y*t : x,3,+/d? -xt,x1 : d?i+/d?;
y,, y,- y*t

i* i

: x,,x,+ /d? - xt,xt- /d? :

:ldlt:of -(/d? - /d?): -(/d? - ad:)

y,, y,- y,, y, : -(/d? - /d?) : O; check

With (13.3)the most important part of (12.10)is then replacedby:

(13.4)

8l

15 CRITERION MATRICES IN NETWORKS FOR

HORIZONTAL CONTROL

Consider the points P, and Pj with referenceto the base points P,,P":
o

( 1s.1)

or with (2.12):

/'l'"):-40O"'l
zi"-

(rs.2)

/:';n:-t#*'"1
In orderto simplifythe notation,

(l 5.3)

the superscript0 will be omitted.

Fig. 15-1

If we introducethe notation:
Re{a.b}:*(a.b+a'.b'1
It can be derivedfrom (15.1)that*):
r) Cf. note to section 15.

(15.4)

puBLIcATIoNsoN GEoDEsy,NEw sERIEs,vor,. 5, No.

92

Case B, see Fig. l5-6

Fig. 15-6

fr

y : ( t r + ' '" '2

ar--as+7t;
l,ili"*lrili":

liil,r,

'

-l l :2 - '' ,

hence:

cosd:0

(15.41)

Qr* Q": Grj"* G,ir- Girj-Gr"j : 0

In both casesthe equations (15.35) are dependent, and one of the equations (15.36) is
lost. The question is if for these casesthe solution for co can be consideredas a limit of

(rs.38).

Case A: This case of a quadrilateral inscribed in a circle is completely determined by

6'0, hence sin d --+0, so that {q,- q,} will approach zero, dependenton sin 6.
Put therefore, withf n a real function:

n,-q":fn'sin6
Thenfrom (15.37):

: +"-tr{3'+-5- t f^\
",,,]30'
Q,: e"

. (rs.42)

c o s d: 1

fahas been left unspecified,but can be written out via formulae as indicated in the note to
this section.

case B: Here, the situation is entirely determinedby cos6--0, so that (4,+4") will
approachzerodependenton cos6.
Put therefore:
q,+ q" : f". cos6
Thenfrom (15.37):

97

S.TRANSFORMATIONS AND CRITERION MATRICES

"iai,i"Ilr".

- Llir,-t""'
+ 1,,!:t'
'"
lr,
1," Ii,

e'nnreo?,, :

11rr.gi(a*i-aui

- gia*!"rlT,

I j'

1,""-in",r,

l r"Ir, ein"'

lr"

1,"

glrt1"II,'i -

gilt,t-"1r",1 -

"nw(l-elr*)
: l-g-iltt.-1

-gnatagll""t,

+e-IIi-

"rlv1r

-t)i,;I)Tiilri;]

ui,;u,"il,i{,

*
nrj

t-_--t
l -o " t-j r
I

+ 1' ,", " ' o " t lrj

:
"iat!"Ilrsr

eior;err",

henCe:

II,s1-

er"nteo?,,: uiri :
:

hence:

or:
'rj

l r r lj " e i n ' " J

,"riei(ni"J*n*,)--

Ir"

lrj
/i,
aio.,y,

henCe:

lrs

- eidtrtens.i -

lirl i" N

lr"
nr* + rrl

/"j

(q + q"si - qn J')
:
Ur sil)r s jei tri

J:

"icr';"

{tstJs e i(d.,., d,,r)

I,"lr"

but this resultis not suitablefor inclusionwith the othereightresults.A differentway goes
via the relationsbetweencoemcients
in (15.52):
l-en*, -gII*r :6
l- en*J_en"', : 0,
hence:
(1 -str^' -su")

(1 - eulr-euli)

: o

or, the sum of the nine coefficients

in (15.53)is zero.Or:

"1*
"ia1'1"11
u1
"ix

IIr"i :
0 - (the other eight results)

t i IIr,

(q
cpsl- ao'.1)
or sil)sr jei r,J*

I)"si1)"rj'-i(n fsr+ds,r) -

"rl,s

l-xli,
hence:

l,"lr" "n.*,

ll3

S-TRANSFORMATIONS AND CRITERION MATRICES

B-conditions for consistent quadrangle

Fig.16-2

Require, apart from two l-conditions for each triangle, per quadrangle ijrs at most two
B-conditions, usually related to the choice of the computational base rs:
Azfit : lztn -:!i

/z[")

(16.10)

-7t t , -

/2{.'it_ /21",)_?

Iru.,,

(16.11)

tsj

(16.10) gives a relation between the triangleTii, jrs and irs, not triangle is7;
(16.ll) gives a relation between the triangles isj, isr and.isr, not trianglejri;
or (8,) and (8") cannot be derived from each other by means of l-conditions.
But .B-conditionscan be transformed by means of ,4-conditions,for example, with:

/:Y': -?,0:l;"
fJ

/:f': -?U"
we obtain:

(16.10):/"['i>: /"['n - ? o"yu

(16.12)

-?o:r*,
(16.11):
*u," : /:fu ' s i

(16.13)

-fJ

But a further transformationis alsopossible,for example,with:

S.TRANSFORMATIONS AND CRITERION MATRICES

t29

Contributionto the conditionmodel,in S :

s.w

Z r s , s ' ) ( " ' - )_ 7 - t + , + t' < " ' - t

"

"

Combination of (17.12)and (17.13)back to S :

'1:'8'gll]
E'rl,;;;

(17.1s)

Combination
of (17.15)with (17.7)backto S :

(ur"'
:,1
"
)
i**i,^,',',,'
ts':;i'
l:'"

(17.16)

l- I x:,,,,,)
lf;::,,,

Combination of (17.16)with (17.3)back to

,S,:

(l',lli:']:"'"''E
(r7.r7)

l/"r','', I

|.Z*"'
)

lo,,io

LZ"")

Then contributionto the conditionmodel,in

f,71t,t'1r,
"', : Lt+,+"1<,'t'>

,S,

(t7.t4")

S-TRANSFORMATIONS AND CRITERION MATRICES

NETHERLANDS TRIANGULATION NETWORK

Observation variates:
directions r (no correlation)
Standarddeviation:o, : 1.1dmgr
Average side length : I x 25 km
Number of observations;q: 449
Number of condition equations:
b: t9l
Number of network points: z : 85

AOJUSTMENT
FROM
STANOARO
ELLIPSE
IN S -SYSTEM
OFNETVORK.
ELLIPSI=CIRCLEFRO.{CRIIERILI4
STAiARO

rlrntx vtrs c.-orof/xu

offis

scALENETVoRK

ooN.

scALE sTAtloARDELLIPSES

Fig. 18-3

t45

s
o
c.l

g , t
9
, &
?F-':

.h6
t r
J
L

>

'-x

s[;
3t
iIe

d
-

tt

l
s
E . t

!t
;lin

_ -

*[

E
9

o
E

jjf.,

z e

.=,

o
E

t r . F

x
ql

V
It
l

E
{

rlr
d
;
-

;:i

8,

E
8,

=
O

XE E
xr so o s
E ;r

F
- F
f t l Z P -

Q
o + V
H
=

v
ill

>

8
E
I

;I A iii IX Es E
C
r- 0i>trl

ril
fJr